Generalized partial response signalling and efficient MLSD using linear Viterbi branch metrics

This paper demonstrates the performance gain attained when noninteger ideal values used by the Viterbi (1967) algorithm (VA) are employed to combat noise correlation and channel nonlinearity in high-density magnetic recording. The noninteger ideal values are determined by the simultaneous adaptation of the equalizer coefficients, target coefficients and the ideal values themselves. The second part of the paper focuses on noninteger VA implementation. Noninteger ideal values in the VA make hardware realization of the VA complex because multiplication operations are required to compute the Euclidean distance branch metrics. A piecewise linear approximation to the Euclidean distance metric is introduced that simplifies realization of the VA by replacing the multiplications with digital left/right bit shifts and add operations. Performance of both the adaptive detection scheme and the piecewise linear approximation are assessed using data collected from a spin stand